Research map / Reinforcement learning / Imitation learning
Action Inference by Maximising Evidence: Zero-Shot Imitation from Observation with World Models
Xingyuan Zhang, Philip Becker-Ehmck, Patrick van der Smagt, Maximilian Karl
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use AIME world-model action inference instead of BCO(0) inverse dynamics model | imitation from observation · V-D4RL random walker | reward/10: 2.11 -> 12.36 | Improved |
| Use AIME instead of BCO(0) | imitation · V-D4RL cheetah-mix | reward/10: 16.08 -> 40.27 | Improved |
| Use AIME instead of BCO(0) | imitation · V-D4RL cheetah-medium_replay | reward/10: 15.47 -> 31.47 | Improved |
| Use AIME instead of BCO(0) | imitation · V-D4RL distracting walker-total_mix | reward/10: 2.15 -> 12.66 | Improved |
| Add visual distractions to embodiment data | AIME · V-D4RL distracting Walker | reward/10: 12.66 -> 3.81 | Hurt |
| Remove KL term or reconstruction term from the ELBO objective (ablation) | AIME · mix -> run, Walker | normalised return | Hurt |
| Use only the KL term instead of both terms (ablation) | AIME · mix -> run, Walker | normalised return | Mixed |
| Remove the state-space model (IIDM: forward model on raw observations) (ablation) | AIME · mix -> run, Walker | normalised return | Hurt |
| Train an explicit IDM on the latent state instead of implicit gradient-based IDM (ablation) | AIME · mix -> run, Walker | normalised return | Hurt |
| Increase number of demonstrations beyond 100 (ablation) | AIME · mix -> run, Walker | normalised return | Improved |
| Use AIME instead of BCO(0) with few demonstrations (2 trajectories) | imitation · mix -> run, Walker | normalised return | Improved |
| Use full-state MDP observation instead of LPOMDP for AIME | AIME · Walker/Cheetah | performance | Hurt |
Closest papers by meaning
- Zero-Shot Offline Imitation Learning via Optimal Transport (ICML 2025 · similarity 0.90)
- Imitation Learning from Observation with Automatic Discount Scheduling (ICLR 2024 · similarity 0.88)
- OSVI-WM: One-Shot Visual Imitation for Unseen Tasks using World-Model-Guided Trajectory Generation (NeurIPS 2025 · similarity 0.88)
- Mimicking Better by Matching the Approximate Action Distribution (ICML 2024 · similarity 0.88)
- Reward-free World Models for Online Imitation Learning (ICML 2025 · similarity 0.88)
- EvIL: Evolution Strategies for Generalisable Imitation Learning (ICML 2024 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
